The Four Genes That Rewind Biological Time

Reversing the Irreversible

For most of the twentieth century, biology was certain a cell’s life ran in one direction only.

A fertilized egg divides into cells that can become anything. Those daughters commit to broad lineages, then to narrower ones, until each lands on a final job — a neuron firing, a skin cell sealing a wound, a white blood cell hunting pathogens. Along the way the cell’s options collapse. And the prevailing assumption, which traces to nineteenth-century ideas about a barrier between the body’s cells and its reproductive ones, was that they collapse for good.

The embryologist Conrad Waddington gave this the image that stuck. In his 1957 book The Strategy of the Genes, he drew a ball perched at the top of a hilly terrain, rolling downward through branching valleys. The ball is a cell. Each valley it enters is a developmental decision. The ridges between valleys are barriers it cannot cross. Reach the bottom and you are a fully differentiated cell, settled in your valley for good. Waddington — who had coined the word “epigenetics” back in the 1940s to describe how genes build a body — meant the picture as a metaphor for how development channels cells toward stable, durable fates.

It hardened into something stronger: an axiom that the ball never rolls back uphill.

Differentiation was terminal. A skin cell stayed a skin cell, and its descendants stayed skin cells, forever.

But this left a genuine puzzle, one you can feel coming the moment you sit through an intro genetics course. Nearly every cell in your body carries the same genome — the same chromosomes, the same DNA sequence, copied faithfully at each division. Your neurons and your liver cells are not different because they hold different genes. They are different because they use different genes. What distinguishes them is the epigenome: the layer of chemical marks and protein packaging laid over the DNA that decides which genes are switched on and which are silenced, without touching a single letter of the underlying sequence. Differentiation, in this view, is just the progressive setting of those switches. So the dogma of irreversibility was really a claim about the epigenome — that once a cell’s switches were set for “skin” or “muscle,” they were locked for life.

The first crack came from frogs, and from John Gurdon. In 1962, working at Oxford, Gurdon ran an experiment in nuclear transfer — you pull the nucleus out of one cell and drop it into another. He took the nucleus from the intestinal cell of a Xenopus tadpole, a cell already committed to its gut-lining fate, and injected it into an unfertilized frog egg whose own nucleus he had inactivated with ultraviolet light. If differentiation were truly irreversible, that transplanted nucleus should have been able to build one thing and one thing only: gut. Instead, some of these reconstructed eggs grew into swimming tadpoles, and a few went all the way to adult frogs — whole animals, every tissue, built from the genome of a single specialized cell.

Gurdon had shown two things at once. A differentiated cell keeps its entire genome intact, throwing nothing away. And the egg held some mixture of factors potent enough to wind that genome all the way back to the start.

What were those factors? Nobody knew. The egg was a black box that worked.

Decades of cloning extended Gurdon’s result to mammals without ever opening the box. In 1996 came Dolly the sheep, out of the Roslin Institute — grown from the nucleus of an adult mammary-gland cell, the only live lamb among 277 attempts. Dolly proved a fully differentiated mammalian cell could still seed an entire organism.

But cloning, by its very design, still needed an egg. The reprogramming happened inside the egg’s cytoplasm, driven by whatever uncharacterized machinery Gurdon had glimpsed. To rewind a cell, you still needed one of nature’s most specialized cells to do it for you.

In 2006, Kazutoshi Takahashi and Shinya Yamanaka, working in Kyoto, threw out the egg entirely. Their reasoning was simple. If the egg’s reprogramming power came from specific proteins, the likeliest suspects were transcription factors — proteins that bind DNA and flip whole sets of genes on or off — that are active in early embryonic cells and silent in adult ones. So they drew up a list of 24 candidate genes, delivered them into adult mouse fibroblasts (ordinary connective-tissue cells, the textbook picture of full differentiation) using retroviruses, and watched for cells that turned embryonic. A tiny fraction did. Then they withdrew the genes one at a time to see which removals broke the effect, and pared the list down to four: Oct3/4, Sox2, Klf4, and c-Myc — the quartet now abbreviated OSKM.

Four genes, dropped into a skin cell, drove it back up Waddington’s landscape to an embryonic-like state that could become any cell type in the body. Yamanaka called the products induced pluripotent stem cells, or iPSCs — “pluripotent” because they can form every cell type, “induced” because the state was forced onto an adult cell instead of inherited from an embryo. The next year, his lab and others pulled it off with human cells.

The radicalism is easy to miss if iPSCs already sound routine to you. Cloning rewound a cell by bathing its nucleus in the egg — a vast, mysterious reservoir of molecules. Yamanaka rewound a cell with four defined genes and nothing else. No egg. No nuclear transfer. No surgery. No borrowed cytoplasm. He had taken the egg’s black box, identified four of the things inside it, and shown those four were enough on their own.

Reprogramming went from a feat that demanded the single most extraordinary cell in biology to a recipe you could write down, hand to another lab, and watch work on the first try. The forces that climb the landscape were not some emergent property of the egg. They were four molecules, and now they had names.

And here’s what’s really interesting. The induced pluripotent stem cell and the fibroblast it came from have identical genomes — same DNA, same sequence, letter for letter. Nothing in the genetic code was edited, added, or removed. The only thing that changed was the epigenome: the pattern of switches was wiped back toward its embryonic setting, and genes silenced for the cell’s entire adult life switched on again.

The four factors did not rewrite the hardware. They reinstalled the operating system.

In other words, cell identity is not a destination written into the genome when development ends. It is a state — a self-sustaining configuration of gene expression — and states can be changed. The ball rolls uphill if you push the right four points.

In 2012 the Nobel Prize in Physiology or Medicine went jointly to Gurdon and Yamanaka “for the discovery that mature cells can be reprogrammed to become pluripotent” — the two ends of a single fifty-year argument. Gurdon, in 1962, proved reprogramming was possible and that an intact genome survives differentiation. Yamanaka, more than four decades later, reduced it to four factors and showed it could be done with no egg at all. Between them they overturned a foundational rule of biology: that the differentiated state, once reached, is final.

Which raises the question this report exists to chase. If four factors can reset the developmental clock — erasing the epigenetic record of having become a skin cell — can they reset the aging clock too?

Aging leaves its own marks on the epigenome. The chemical modifications on our DNA shift in measurable, strikingly predictable ways as we get older — regular enough that biologists now read them as an “epigenetic clock.” Differentiation and aging both look like one-way trips engraved in the epigenome, and Yamanaka shattered the first taboo against running one of them backward.

The obvious next move is to push the factors only partway — rejuvenating a cell without erasing what it is. That has already made aged tissues in mice behave young again. Whether the same logic can reach human aging is the wager now driving an entire field.

From Twenty-Four to Four

The story begins not with a stem cell but with a frog.

In 1962, John Gurdon was a young researcher at the University of Oxford. He ran an experiment that demolished a deep assumption about how cells specialize.

The reigning view went like this. As a cell matures into a particular type — a skin cell, a gut cell — it permanently discards or silences the genetic instructions it no longer needs. Differentiation is a one-way street, paved by the loss of genomic information.

Gurdon tested the idea head-on. He took the nucleus — the compartment that houses a cell’s chromosomes, and therefore its full genome — out of a differentiated intestinal epithelial cell of a feeding Xenopus tadpole. Then he injected it into a frog egg whose own nucleus he had destroyed with ultraviolet light.

If the gut cell’s genome had really been stripped down, the egg should go nowhere.

It didn’t. A small fraction of these reconstructed eggs developed normally. Of the first batch, roughly ten in 726 grew into swimming tadpoles, and some went on to become adult frogs. And when he repeated the transfer serially — taking nuclei from a developing clone and transplanting them again — the success rate climbed higher still.

The conclusion was inescapable. A differentiated cell keeps its entire genome, intact and complete.

This is the principle of genomic equivalence. The difference between a neuron and a liver cell is not written into their DNA sequences — those are essentially identical — but into which genes are switched on and which are switched off. And that pattern of switches, Gurdon had shown, could be wiped clean. Something in the egg’s cytoplasm, the crowded fluid surrounding the nucleus, could reprogram a mature nucleus all the way back to an embryonic state.

The same feat was later pulled off in mammals. Most famously in 1996, when Dolly the sheep was cloned from the nucleus of an adult mammary cell.

What Gurdon had not shown was how.

The reprogramming machinery was hidden inside the egg — an intricate broth of thousands of proteins, RNAs, and small molecules. No one knew which of its ingredients did the work.

Enter Shinya Yamanaka. Working at Kyoto University, he reframed the question in a way that made it tractable. Instead of asking what a whole egg does, he asked something far narrower: could a small, defined, namable set of factors — molecules you could clone and add to a cell on purpose — pull off the same reset, with no egg involved at all?

His hypothesis was elegant in its symmetry. Embryonic stem cells (ESCs), harvested from the early embryo, are pluripotent: they renew themselves indefinitely in culture and can give rise to every tissue in the body. A particular set of genes holds them in that state, keeping the pluripotency program switched on and the various differentiation programs switched off.

Here was Yamanaka’s leap. The same factors that maintain pluripotency in an embryonic stem cell might also be enough to induce it from scratch. Force those factors into an ordinary, fully committed cell, and you might override its identity and drag it back toward a stem-cell-like state. Maintaining the program and installing it, in other words, might be two faces of one molecular toolkit.

This was not an obvious bet. Many assumed differentiation in mammals was effectively irreversible without an egg’s help.

The first job was to name the suspects. Yamanaka and his graduate student Kazutoshi Takahashi drew up a list of 24 candidate genes, each one specifically active in, or important to, embryonic stem cells — what the lab called ES-cell-associated transcripts.

The roster was eclectic. There were transcription factors — proteins that bind DNA and switch genes on or off — characteristic of ESCs, such as Oct3/4, Sox2, and Nanog. There were growth- and cancer-related genes that matter in stem cells, such as c-Myc and Klf4. And there was a tail of more obscure ESC-enriched genes, among them one called Fbx15.

To get these genes into cells, the team used retroviruses, which permanently splice their genetic cargo into a host cell’s chromosomes. The target was mouse embryonic fibroblasts — the flat, unremarkable connective-tissue cells that are a standard laboratory workhorse.

The real obstacle was detection. If the scheme worked at all, success would be rare — a handful of cells among many thousands. The lab needed a way to make those rare events announce themselves.

The trick relied on Fbx15 itself, which is switched on in embryonic stem cells but silent in fibroblasts. Using homologous recombination — a technique for swapping a designed piece of DNA into a precise spot in the genome — Yamanaka had earlier engineered mice in which the Fbx15 gene drove a cassette called βgeo, a fusion that, among other things, confers resistance to the antibiotic G418.

The logic falls out cleanly. In an ordinary fibroblast, Fbx15 is off, no βgeo is made, and the cell dies when you dose it with G418. But if a cell’s pluripotency program is genuinely reactivated, Fbx15 switches on, βgeo with it, and the cell survives the drug. Selection does the searching for you: flood the dish with G418, and only the cells that have been successfully reprogrammed live to grow into visible colonies.

Takahashi introduced all 24 factors at once and applied selection. The payoff appeared: drug-resistant colonies of ES-cell-like cells rising out of the fibroblast lawn after about three weeks.

Then he tried each factor on its own. Nothing. No lone gene could flip the switch.

So some combination within the 24 was responsible. But the number of possible combinations was astronomical, and testing them by addition would have taken forever. Here was the inspired move, and it was a study in subtraction rather than construction. Take the full set that works. Remove factors one at a time. See what breaks. Drop a gene and nothing changes, it’s dispensable. Drop a gene and the colonies vanish, it’s essential.

Round by round the list shrank. First to a core of ten genes that still produced colonies. Then, by the same merciless logic, to an irreducible four. Withdraw any one of them and reprogramming either failed outright or collapsed to a trickle. Supply all four and it proceeded.

The four were Oct3/4, Sox2, Klf4, and c-Myc — the combination the world would come to call the Yamanaka factors.

In August 2006, Takahashi and Yamanaka published the result in Cell and named the new cells induced pluripotent stem cells, or iPS cells. This was the conceptual watershed. Four defined proteins, not the inscrutable interior of an egg, could turn back a cell’s developmental clock.

The first iPS cells were far from perfect. The ones picked out by Fbx15 selection looked like embryonic stem cells but weren’t fully equivalent — they differed in their gene-expression profiles and in their DNA methylation patterns, the chemical tags on DNA that help keep genes locked off, and they flunked the most demanding tests of true pluripotency. But the principle was secure. And improvements came fast.

The pivotal question was whether the same magic would work on human cells. That’s where the payoff — modeling disease, and someday growing replacement tissues — was vastly greater.

In late 2007 the answer arrived twice, nearly at once, from opposite sides of the Pacific.

Yamanaka’s group, again in Cell, reprogrammed adult human dermal fibroblasts with the very same four-factor recipe: OCT4, SOX2, KLF4, and c-MYC. That same month, in Science, a team led by Junying Yu and James Thomson at the University of Wisconsin–Madison reported human iPS cells by an independent route, using OCT4, SOX2, NANOG, and LIN28.

The two recipes shared a core — both leaned on OCT4 and SOX2 — but they diverged at the edges, and the differences are instructive. NANOG had been on Yamanaka’s original list of 24, yet it hadn’t survived his subtraction. LIN28 wasn’t a transcription factor at all but an RNA-binding protein, one that governs which messenger RNAs get translated into protein — the only non-transcription-factor among the first reprogramming genes.

Think about what this means. Two labs, setting out from different candidate sets, converged on overlapping recipes for the identical outcome. That’s strong evidence that pluripotency could be installed on purpose.

The genome’s developmental memory — as Gurdon had first hinted with his frogs nearly half a century earlier — was erasable by design. The Nobel committee said as much in 2012, when it awarded Gurdon and Yamanaka the prize in Physiology or Medicine together.

Meet OSKM

The 2006 mouse experiment that launched the field didn’t start with four genes.

It started with twenty-four.

Kazutoshi Takahashi and Shinya Yamanaka drew up a list of genes known to be active in embryonic stem (ES) cells, or implicated in keeping those cells unspecialized. They loaded all of them into adult mouse skin cells (fibroblasts) using retroviruses — viruses engineered to splice their cargo permanently into the host genome — and then pulled candidates out one at a time to see which ones actually mattered.

The pool collapsed to four: Oct4, Sox2, Klf4, and c-Myc, or OSKM. All four are transcription factors, proteins that bind specific DNA sequences to switch nearby genes on or off — though one of them, as we’ll see, acts more like an amplifier than a switch. So why this particular quartet, and not the dozens of other plausible candidates? To answer that, you have to look at what each protein does for a living.

Oct4, encoded by the gene POU5F1, is the linchpin. It belongs to the POU family, named for a shared DNA-binding module, the POU domain, that reads an eight-letter sequence called the octamer (ATGCAAAT) by gripping the major groove of the double helix. In the early embryo, Oct4 marks exactly the cells that still hold full developmental potential. Lose it, and those cells differentiate or die.

But the real reason Oct4 matters is that it’s a pioneer factor. Most transcription factors can only dock onto DNA that’s already exposed. The problem is that DNA in the nucleus is wound around spools of histone proteins called nucleosomes — the packing unit that crams roughly two meters of genome into a microscopic nucleus — and most of it is folded away, out of reach. A pioneer factor is one of the rare proteins that can grip its target sequence even while it’s wrapped onto a nucleosome, prying open stretches of chromatin (the DNA-plus-protein material of the chromosome) that ordinary factors can’t touch.

Of the four, Oct4 is the hardest to replace. The nuclear receptor Nr5a2 can stand in for the added Oct4 in mouse reprogramming, but only by switching on the cell’s own dormant Pou5f1 gene — the route still runs through Oct4. Even fully chemical recipes end up reactivating the endogenous copy. Whether you supply it from outside or coax it from within, Oct4 protein is non-negotiable. No working cocktail omits it.

Sox2 is Oct4’s partner. Its name comes from “SRY-related HMG box,” after its own DNA-binding module, the high-mobility-group (HMG) domain, which clamps onto the minor groove of the helix and bends the DNA sharply. The two proteins work as a single physical unit. They dock side by side on composite “sox–oct” motifs — an HMG site (something like CATTGTC) sitting right next to an octamer — that stud the control regions of pluripotency genes like Fgf4 and Nanog. The spacing between the two sites tunes how tightly they cooperate; in the best-studied case, the Fgf4 enhancer, it’s exactly three nucleotides.

Sox2 is a pioneer factor too, and the Oct4–Sox2 pair opens chromatin that neither one reaches as well alone. So Sox2 isn’t a redundant backup for Oct4. It’s the complement — reading the opposite groove of the same DNA and stabilizing Oct4 on the partner protein.

Klf4 — Krüppel-like factor 4 — is a zinc-finger transcription factor: it grips DNA with small looped protein modules, each pinched into shape by a zinc ion. (The family is named for its resemblance to the Krüppel gene of the fruit fly.) In the body, Klf4 concentrates in the mature, non-dividing cells lining the gut and skin, where it enforces epithelial identity. Two of its jobs earn it a seat in the cocktail.

First, it pushes cells toward that epithelial state. A fibroblast is a mesenchymal cell — loosely attached, migratory — whereas ES cells grow as tightly bound sheets. So reprogramming has to run through a mesenchymal-to-epithelial transition (MET) early on. Klf4 drives it by switching on E-cadherin (the CDH1 gene), the adhesion protein that glues epithelial cells together, cranking its expression up more than two-hundred-fold.

Second, Klf4 blocks apoptosis — programmed cell death — which the stress of reprogramming would otherwise trigger. It does this by repressing the tumor-suppressor protein p53 and the pro-death gene BAX.

And that same anti-death wiring is exactly what makes Klf4 dangerous. It’s a context-dependent gene. In some tissues it acts as a tumor suppressor, stopping the cell cycle after mild DNA damage. In others it’s an oncogene — a cancer-driving gene — keeping damaged cells alive and dividing. The very property that lets a fibroblast survive its conversion can, in the wrong setting, let a tumor cell survive too.

c-Myc is the odd one out. It’s a proto-oncogene — a normal gene whose overactivity drives cancer — and in the cocktail it’s the accelerator, not the steering wheel. Instead of going after a focused set of pluripotency genes, Myc acts broadly across the genome. It ramps up proliferation and rewires metabolism toward the fast, glycolysis-heavy fuel program that rapidly dividing embryonic cells run on. Left to itself, the OSK trio hits a proliferation pause enforced by the retinoblastoma (RB) protein. Myc overrides that brake in the first few days, so cells divide fast enough to finish the conversion.

The payoff is efficiency. With Myc, far more colonies appear, and faster.

The cost is cancer. In Yamanaka’s 2007 follow-up, roughly 20% of the mice made from iPS cells developed tumors, and 24 of 121 offspring of the chimeric mice died of tumors traced back to reactivation of the silenced c-Myc retrovirus. And because that retrovirus is wedged permanently into the genome, it can flip back on long after reprogramming is done — a latent oncogene riding along in every descendant cell.

That reactivation risk kicked off a search for safer formulations. And it turned out none of the four factors is truly sacred — except Oct4.

In 2008, Yamanaka’s group made iPS cells from just OSK, dropping Myc entirely. The cells came far more slowly and in much smaller numbers, but the mice carried far fewer tumors. A better fix came in 2010, when the same lab swapped in L-Myc, a Myc relative with almost no transforming activity. It reprogrammed as efficiently as c-Myc, or better, yet the chimeric mice passed the cells through the germ line without tumors.

Meanwhile, James Thomson’s group in Wisconsin — which reported the first human iPS cells within weeks of Yamanaka in late 2007 — used a completely different quartet: OCT4, SOX2, NANOG, and LIN28, the so-called Thomson factors. Nanog is itself a core pluripotency transcription factor, and notably one of the original twenty-four that the mouse screen had thrown out as dispensable. Lin28 isn’t a transcription factor at all. It’s an RNA-binding protein that blocks maturation of the let-7 family of microRNAs (short RNAs that silence their targets), which lifts a brake on stemness and proliferation. The factors behave like partly interchangeable building blocks, not a fixed recipe.

The chemists pushed this logic the furthest. In 2013, Hongkui Deng’s lab in Beijing made mouse iPS cells using only seven small molecules and no added genes, at a frequency around 0.2% — comparable to the gene-based method — with healthy, fully viable chimeras. By 2022 the group extended purely chemical reprogramming to human cells. And even Oct4 gives ground here, because the small molecules end up reactivating the cell’s own Oct4 gene instead of supplying the protein directly.

The real lesson is that the cocktail works as a system, not as a bag of individually decisive genes. No single member of OSKM makes a stem cell.

Oct4 and Sox2 pry open silenced regulatory DNA as pioneer factors and bind cooperatively at shared composite sites. Klf4 reinforces them, drives the epithelial transition, and keeps the stressed cell alive. Myc clears the proliferative and metabolic roadblocks so the conversion can finish in a reasonable amount of time. Acting together, they switch on the cell’s own pluripotency genes — its native Oct4, Sox2, and Nanog — and those genes sustain the state once the introduced factors fade away.

This is what explains both halves of the story: why twenty-four candidates collapsed to four, and why the four tolerate so many substitutions. Pluripotency is a self-reinforcing genetic network. The cocktail’s only real job is to shove the cell across the threshold where that network switches itself on and locks in place.

In other words, the specific identity of any one factor matters less than getting the full set of jobs done at once, in the same cell: open the chromatin, pair at the right motifs, convert to an epithelial state, suppress cell death, supply proliferative drive.

Inside the Reprogramming Machine

Every cell in your body carries the same genome. Yet a skin fibroblast and an embryonic stem cell could hardly behave more differently.

Why?

It’s not which genes a cell has. It’s which genes it switches on. And that pattern is enforced by how tightly the DNA is packaged. Inside the nucleus, DNA is wound around spool-like clusters of proteins called histones. One spool — roughly 147 base pairs plus its histone core — is a nucleosome, and long runs of nucleosomes fold into dense, inaccessible fibers. Where this chromatin — the DNA-plus-protein complex — is condensed, the genes inside it are effectively hidden from the machinery that would read them.

So a cell’s identity is written twice. Once in the transcription factors it produces, and once in the chromatin landscape that decides which genes those factors can even reach. To turn a fibroblast into a pluripotent cell, the four Yamanaka factors have to rewrite both layers. And they do it in a surprisingly ordered sequence.

Pioneer factors pry open closed chromatin

The central obstacle is access.

The genes that define pluripotency — Oct4, Sox2, Nanog, and their partners — sit in chromatin that, in a fibroblast, is shut tight. Most transcription factors can only bind DNA that’s already exposed. Nucleosomal DNA locks them out.

Three of the four reprogramming factors are the exception. Oct4 (which grips DNA through a structure called a POU domain), Sox2 (an HMG-box protein), and Klf4 (a zinc-finger protein) are pioneer transcription factors — proteins that can find and bind their target sequences even when those sequences are wrapped around a nucleosome. Abdenour Soufi and Kenneth Zaret’s group showed this directly. Purified Oct4, Sox2, and Klf4 bind nucleosomes in the test tube. And when first introduced into human fibroblasts, they home in on exactly the chromatin ordinary factors can’t touch: predominantly silent, nucleosome-rich, nuclease-resistant — the closed regions.

How do they pull it off? They relax their demands.

A nucleosome surface exposes only part of any wrapped sequence at a time. So instead of the complete binding sites they’d use on naked DNA, the pioneers recognize partial versions of their motifs — shorter, looser stretches that tolerate weaker contact with the DNA. Oct4, for instance, tends to dock near the points where the DNA enters and exits the histone spool, where the sequence is most accessible. This initial, low-affinity grip is the foothold. Once anchored, the pioneers recruit chromatin-remodeling complexes — molecular machines like the BRG1-containing remodeler that burn ATP to slide and evict nucleosomes — and pry the local packaging open, so the site becomes available to the broader transcriptional apparatus.

The fourth factor, c-Myc, is not a pioneer. On its own it largely can’t crack open closed chromatin. Instead it binds sites that are already accessible, amplifies the binding of the other three, promotes an active chromatin state, and drives the cells to divide rapidly. That dual role — boosting both factor binding and proliferation — is why c-Myc sharply increases reprogramming efficiency yet remains dispensable. You can make iPSCs without it. Just more slowly, and more rarely.

A trajectory in two phases

Reprogramming is slow and inefficient. In mouse fibroblasts, typically well under one percent of cells complete the journey, and they take one to three weeks to do it.

Single-cell studies explain the mess.

In 2012, Yosef Buganim, Rudolf Jaenisch, and colleagues profiled gene expression in individual reprogramming cells, and found the process splits into two phases. The early phase is stochastic — probabilistic and disorderly. Cells that have taken up the factors look wildly different from one another in which genes flicker on, with no fixed order. The factors engage thousands of sites, most cells stall, and only a minority stumble onto a productive route.

Then the character of the process changes.

A late deterministic, hierarchical phase kicks in, where gene activation follows a fixed, predictable cascade. In the Buganim study, the trigger was the cell activating its own Sox2 gene, which set off an ordered chain of downstream events. Konrad Hochedlinger and colleagues mapped the same trajectory a different way, by sorting cells with surface markers: cells first switch off the fibroblast marker Thy1, then gain the early marker SSEA1, and only much later reactivate their endogenous Oct4. The pluripotency genes come on in waves — Nanog and Esrrb during a maturation phase, others like the cell’s native Sox2 not until a final stabilization phase.

But the first organized event in this trajectory, before the pluripotency network even wakes up, is something else entirely: a wholesale change in the cell’s shape and behavior called the mesenchymal-to-epithelial transition, or MET. Fibroblasts are mesenchymal — loosely attached, migratory, spindle-shaped. Pluripotent cells are epithelial — tightly packed colonies whose members are glued together by adhesion proteins. Two independent 2010 studies, one from Duanqing Pei’s group in Guangzhou and one from Jeff Wrana’s lab in Toronto, showed that this transition both kicks off and is required for fibroblast reprogramming.

The factors orchestrate it directly. Klf4 switches on epithelial genes, inducing the cell-adhesion protein E-cadherin more than 200-fold. Oct4 and Sox2 suppress Snail, a master regulator of the opposite (epithelial-to-mesenchymal) transition. And c-Myc shuts down TGF-β signaling, which would otherwise hold the cells in the mesenchymal state. Push the cells the other way — add TGF-β, or force Snail expression — and reprogramming fails.

MET is the gate. The cells have to pass through it first.

Throughout all of this, two opposing programs run in parallel. The somatic program that makes a fibroblast a fibroblast gets shut down, while the pluripotency network gets switched on.

The crucial word is endogenous.

At the start, all the pluripotency-driving activity comes from the artificially supplied factors — usually delivered by viruses as extra gene copies, or transgenes. But genuine reprogramming requires the cell to activate its own native Oct4, Sox2, and Nanog genes. Those genes then bind one another’s regulatory regions, and their own, forming a self-reinforcing circuit that holds the pluripotent state in place. This is why reactivating the endogenous genes is a comparatively late milestone, and the true mark of success. Cells that display the surface markers but never ignite their own core genes are only partially reprogrammed — stranded short of the goal.

The epigenetic reset

Locking in the new identity means rewriting the cell’s epigenome — the chemical marks on DNA and histones that record which genes should be on or off, without touching the underlying sequence, and that get copied along as the cell divides.

Three kinds of marks get overhauled.

DNA methylation

Methyl groups attached to cytosine bases, usually where a C sits next to a G. They generally silence a gene, and fibroblasts use them to keep the pluripotency promoters off. Reprogramming has to strip these marks off the Oct4 and Nanog promoters. The cell’s own TET enzymes — TET1, TET2, TET3 — do much of the work, oxidizing methylated cytosine into 5-hydroxymethylcytosine, the first step toward erasing the mark. TET2 acts early, modifying genes like Nanog and Esrrb and priming them to fire. This demethylation is essential — so essential that TET1 can stand in for Oct4 in the cocktail entirely, by demethylating and reactivating the endogenous Oct4 gene.

Histone modification

Chemical tags on the histone tails flag the state of nearby genes. Trimethylation of lysine 4 on histone H3 — written H3K4me3 — marks active promoters. H3K27me3 and H3K9me3 mark repressed, compacted chromatin. Reprogramming redraws the map: pluripotency genes gain the activating H3K4me3, and repressive marks get stripped from key sites (the enzyme KDM4C, for instance, removes H3K9me3 from the Nanog promoter to let it switch on). Many developmental genes end up in bivalent domains — carrying an activating and a repressive mark at the same time, poised, neither fully on nor off. That’s a configuration characteristic of stem cells. And alongside these chemical edits, the chromosomes physically refold, rearranging the 3D contacts between distant regulatory elements into the layout of an embryonic cell.

X-chromosome reactivation

This one shows up only in female cells. To balance gene dosage between the sexes, female somatic cells silence one of their two X chromosomes, compacting it into a dense, largely inert mass. Pluripotent cells keep both X chromosomes active. So reprogramming has to reawaken the silenced one — a late and demanding step that, like the activation of the endogenous core genes, signals the cell has reached the bottom of the developmental hill.

The handoff

The finale is a handoff.

For roughly the first ten days, the cell leans entirely on the delivered factors to push everything forward. But as the endogenous circuit comes online and the epigenome gets reset, the cell’s own genes start producing enough Oct4, Sox2, Nanog, and their partners to sustain the pluripotent state on their own.

At that point the cell silences the viral transgenes. It switches off the very factors that launched the process, and becomes transgene-independent.

This is the defining test of a true induced pluripotent stem cell. It no longer needs the four factors, because the self-sustaining circuit they bootstrapped has taken over.

The scaffolding comes down, and the rebuilt cell stands on its own.

The Cell Fights Back

When Takahashi and Yamanaka first turned mouse fibroblasts into pluripotent cells, everyone fixated on the headline and glossed over the awkward part: it almost never worked.

Of all the cells that received the four factors, only a sliver — somewhere between 0.01 and 0.1 percent, and maybe approaching 1 percent under optimized conditions — ever became a genuine induced pluripotent stem cell, or iPSC. The rest either clung to their original identity or died.

And the rare winners were slow. Mouse cells need roughly one to two weeks. Human cells take three to four. During that lag the factor-expressing cells sit in an intermediate limbo, dividing many times before a rare descendant finally crosses over.

Tracking individual cells tells us the conversion is largely stochastic. Most cells carrying the factors could in principle become an iPSC if you followed them long enough — but only after many divisions, and on no fixed schedule, as if each division were another low-probability roll of the dice to slip past the cell’s defenses.

That combination — rare and slow — is the whole clue. Reprogramming isn’t a switch the factors flip. It’s a siege they lay against a cell that actively defends its own identity.

The Guardian of the Genome

Here’s the part people find counterintuitive: the best-studied defenses are the exact same proteins that guard against cancer.

Chief among them is p53, often called the “guardian of the genome.” It’s a transcription factor that senses cellular stress — DNA damage, the strain of forced rapid division, or the abnormal activity of the reprogramming factors themselves — and responds by arresting the cell cycle or triggering programmed cell death. One of its main effectors is p21, which physically blocks the machinery that pushes a cell from one division to the next.

In 2009, in a coordinated burst of papers, five separate groups reported the same thing: this p53–p21 axis is a major brake on reprogramming. Knock down p53 and efficiency jumps dramatically. In one study, up to 10 percent of p53-deficient mouse fibroblasts became iPSCs — and they crossed over faster, and with fewer factors, than normal cells.

The Second Barrier

A second barrier sits at a single, remarkable stretch of chromosome: the INK4a/ARF locus. This one stretch encodes two distinct tumor suppressors from overlapping reading frames — p16INK4a and ARF (p19ARF in mouse, p14ARF in human).

p16 enforces senescence: a permanent exit from the cell cycle, where the cell stays alive and metabolically active but will never divide again. ARF, for its part, stabilizes p53. The reprogramming factors switch this locus on, and the resulting senescence and arrest pull candidate cells out of the running. Delete INK4a/ARF, just like silencing p53, and reprogramming gets markedly more efficient.

Notice that senescence matters doubly here, because it accumulates with age. Cells from older donors carry more p16 and reprogram more poorly. That’s one reason patient-derived lines vary so much.

Why You Can’t Just Rip Out the Guardrails

By now there’s an obvious temptation: tear down these guardrails and manufacture iPSCs at scale. And an equally obvious reason not to.

p53, p21, and INK4a/ARF are among the most important tumor suppressors in the body. p53 is the single most frequently mutated gene in human cancer. The INK4a/ARF locus is among the most commonly deleted. This is no coincidence. The barriers to reprogramming and the barriers to malignant transformation are, to a striking degree, the same barriers — because the two processes share machinery. Both require a cell to escape its differentiated state, divide indefinitely, and tolerate genomic stress.

So a cell stripped of p53 reprograms beautifully. But it also waves through the very mutations and chromosomal abnormalities p53 exists to catch.

And the danger isn’t hypothetical. One of the original reprogramming factors, c-Myc, is a potent oncogene, and early iPSC-derived mice frequently developed tumors when the retrovirus carrying c-Myc reactivated. This is the field’s defining trade-off: efficiency, bought at the price of the safeguards that keep an unleashed cell from becoming a tumor.

The Product Is Never a Blank Slate

Even when reprogramming works, the product is rarely a blank slate.

A cell’s identity is written largely in epigenetic marks — chemical modifications layered onto the DNA and the proteins it wraps around, that switch genes on or off without touching the underlying sequence. The best-studied is DNA methylation: methyl groups attached to cytosine bases, which generally silences nearby genes. Reprogramming is supposed to wipe the donor cell’s methylation pattern and re-impose the embryonic one. In practice it does this imperfectly.

In 2010, Kim and colleagues showed that low-passage iPSCs — lines only recently established, before many rounds of splitting and regrowing the cells — retain residual methylation signatures of the tissue they came from. An “epigenetic memory.” A blood-derived iPSC still carries molecular traces of having been a blood cell.

This memory isn’t cosmetic. It biases what the cell can later become, making iPSCs differentiate more readily back toward their lineage of origin while resisting conversion into unrelated cell types. You can dilute it — with extended culture, repeated rounds of differentiation and re-reprogramming, or drugs that strip chromatin marks — but its existence means two iPSC lines that look identical under the microscope may behave very differently depending on their histories.

Active Damage

Worse than imperfect erasure is active damage. Reprogramming, and the long culture that follows, introduce genuine genetic changes. A wave of studies in 2011 catalogued them.

• Gore and colleagues sequenced the exome — the protein-coding regions — of 22 human iPSC lines made in seven labs by five different methods. They found, on average, about six coding mutations per line that were present in the iPSC but not in the parent fibroblast. These point mutations were enriched in genes implicated in cancer.
• Hussein and colleagues found that iPSCs carry more copy-number variations — deletions or duplications of whole stretches of chromosome — than either the parent cells or embryonic stem cells. The burden is highest early on, then gets pruned over time as fitter cells outcompete damaged ones in culture.
• Laurent and colleagues added that extended passaging specifically favors deletions of tumor suppressors and duplications of growth-promoting oncogenes.

The aberrations have mixed origins. Some pre-exist as rare variants in the starting tissue and are merely captured when a single cell gets reprogrammed. Others are generated by the process itself. Still others arise and get amplified by selection during culture. The result is the same: a given iPSC line may harbor mutations, chromosomal gains and losses, and aberrant methylation that its donor never had.

Is It Even Reprogrammed?

Finally, even deciding whether a colony is reprogrammed at all is harder than it sounds.

Many of the colonies that emerge are only partially reprogrammed — sometimes called pre-iPSCs. They’ve activated some pluripotency markers and changed shape, but never completed the transition. They fail to fully silence the introduced factors, fail to switch on the cell’s own pluripotency genes, and stay stuck in limbo — often nearly indistinguishable from the real thing by eye.

So confirming bona fide pluripotency requires functional tests, arranged in a hierarchy of stringency.

The minimal bar is forming a teratoma — a benign tumor containing tissues from all three embryonic germ layers, the ectoderm, mesoderm, and endoderm that build the entire body — when you inject the cells into a mouse. Even here, partially reprogrammed cells can occasionally sneak through.

The most demanding test, available only in mice, is tetraploid complementation. You combine the candidate cells with an embryo whose own cells have been fused to carry four chromosome sets, so those cells can build only the placenta. Any animal that develops is therefore constructed entirely from the injected iPSCs. Produce a healthy “all-iPSC” mouse and you have the gold standard — proof the cells are fully pluripotent.

The need for assays this elaborate underscores the whole theme. Coaxing a cell to abandon its identity is rare and slow. And even the cells that appear to have done it often carry the fingerprints of the state they never entirely left.

From Virus to Molecule

When Shinya Yamanaka’s team first turned skin cells back into stem cells in 2006, and pulled off the same trick with human cells in 2007, they delivered the four reprogramming factors — Oct4, Sox2, Klf4, and c-Myc — using retroviruses. A retrovirus copies its genetic payload straight into the DNA of the cell it infects, splicing the new genes permanently into a chromosome.

That’s exactly what made the method work. The four factors switched on and stayed on long enough to flip the cell’s identity.

It’s also exactly what made it dangerous.

Here’s the problem. A retrovirus doesn’t pick its landing site carefully. It inserts semi-randomly, and a single reprogrammed cell typically ends up carrying many such insertions scattered across its genome. Each one is a small act of vandalism. Land inside a working gene, and you can break it. Land next to a gene that controls cell growth, and you can switch it on when it has no business being on. This hazard — a foreign sequence wrecking or mis-activating a host gene by sheer bad luck of position — has a name: insertional mutagenesis, the central safety problem of any integrating vector.

James Thomson’s group reported human iPS cells the same week in 2007, using a lentivirus — a retrovirus relative, HIV being the famous example — that carried a different quartet: OCT4, SOX2, NANOG, and LIN28. But lentiviruses share the same liability. They integrate.

There was a second, subtler danger. Reprogramming is supposed to be hit-and-run. The factors do their job, then fall silent as the cell settles into its new pluripotent state. But integrated genes can wake back up. And here’s the elephant in the room: two of the four factors, c-Myc and Klf4, are proto-oncogenes — genes that drive cancer when overactive. In a 2007 study, Yamanaka’s group bred mice from their iPS cells and watched roughly one in five of the offspring develop tumors, traced to reactivation of the inserted c-Myc gene.

The lesson was unambiguous. A cell carrying permanent copies of cancer-linked genes is not a cell you put into a patient.

The campaign to go footprint-free

The decade that followed was, more than anything, one long campaign to deliver the same instructions without leaving a permanent mark.

The first workaround kept a virus but changed its biology. Sendai virus, a respiratory virus of rodents, is an RNA virus that replicates entirely in the cytoplasm and never enters the nucleus, so it gets no chance to splice itself into a chromosome. In 2009 Fusaki and colleagues used it to make iPS cells. As the cells divide, the virus is diluted out and cleared over a few passages, leaving a stem cell with zero genetic footprint. Marketed as CytoTune, it became one of the most widely used reprogramming tools, precisely because it pairs high efficiency with no integration.

Other groups dropped viruses entirely.

One approach used episomes — circular pieces of DNA that float free in the nucleus and copy themselves as independent units instead of joining a chromosome. In 2009 Junying Yu and colleagues built plasmids carrying a fragment borrowed from Epstein–Barr virus, the oriP/EBNA1 system, which lets each circle replicate about once per cell cycle. Because the circles are lost at roughly five percent per division, cells that have shed every last trace of the introduced DNA can simply be picked out later. One transfection is all it needs. The catch is that it reprograms inefficiently.

A more elegant DNA trick used a transposon — a “jumping gene” that can hop into and out of the genome. The piggyBac transposon integrates like a virus, but it can be excised cleanly afterward: a brief pulse of the transposase enzyme cuts it back out and restores the original sequence exactly, with no scar left behind. Two 2009 Nature papers, one from Kaji and one from Woltjen, showed reprogramming followed by exactly this seamless removal.

The purest nucleic-acid approach skipped DNA altogether and delivered messenger RNA — the short-lived working copy a cell normally transcribes from a gene to build a protein. mRNA never becomes DNA, so it cannot integrate. It is translated, used, and degraded within days. The catch was that cells treat foreign RNA as a sign of viral infection and mount an innate immune response — a blunt anti-pathogen alarm — that shuts protein synthesis down. In 2010 Luigi Warren and colleagues got around this by chemically tweaking the RNA’s building blocks, swapping in pseudouridine for uridine and 5-methylcytidine for cytidine, which let the synthetic transcripts slip past the cell’s RNA sensors. Repeated daily transfections of this modified mRNA reprogrammed human cells efficiently and left nothing behind.

Most cautious of all was delivering the factors as finished proteins, tagged with short stretches of arginine that ferry cargo across the cell membrane. Zhou and, separately, Kim reported protein-induced iPS cells in 2009. Conceptually it’s the safest method of the lot — no genetic material of any kind enters the cell. In practice it barely works: on the order of 0.001 percent of treated cells, versus roughly 0.01 to 0.1 percent for viruses, and the bulky reprogramming proteins are a nightmare to manufacture and purify in quantity.

Three axes

These methods trade off against one another along three axes: efficiency, safety, and manufacturability.

Integrating viruses are efficient but unsafe. Protein delivery is safe but barely works. Between those poles, Sendai virus, episomes, and modified mRNA emerged as the practical winners for clinical work, because each is non-integrating and each can be produced under Good Manufacturing Practice — the documented, reproducible standards regulators demand for anything destined for a human body.

Ditch the factors entirely

Then came a more radical idea: get rid of the transcription factors altogether. If the cell’s own dormant pluripotency genes could be coaxed awake with drugs, then nothing foreign — no virus, no plasmid, no RNA, no protein — would need to enter the cell at all.

In 2013 Hongkui Deng’s lab in Beijing reported chemically induced pluripotent stem cells, or CiPSCs: mouse cells reprogrammed by a cocktail of small molecules, not a single transcription factor among them, at an efficiency as high as 0.2 percent. The mix, which they abbreviated VC6TFZ, combined drugs that loosen the cell’s tightly packed DNA and rewire its internal signaling — valproic acid, for instance, which blocks the enzymes that keep stretches of DNA compacted and silent, and CHIR99021, which inhibits the GSK3 signaling kinase. Forskolin, a molecule that raises the second messenger cAMP, turned out to substitute for Oct4 itself.

A small molecule standing in for a master transcription factor.

Getting the trick to work in human cells took nearly another decade. In 2022 the Deng lab reported chemically reprogrammed human cells, by way of a longer, staged protocol — roughly 50 days, with small molecules added in sequence. First to erase the cell’s original identity and push it into a flexible, “regeneration-like” intermediate state, then to switch on the pluripotency network. A key insight was that the JNK stress-signaling pathway acts as a barrier to plasticity and has to be suppressed. Efficiencies climbed to a few percent — around 0.2 to 2.6 percent, comparable to factor-based methods — using nothing but defined, off-the-shelf chemicals that can be weighed out, dosed, and washed away.

The price of admission

Why did all this engineering matter?

Because integration-free, footprint-free reprogramming turned out to be the price of admission to the clinic. When RIKEN performed the world’s first iPS-cell transplant in 2014 — a sheet of retinal pigment cells grown to treat age-related macular degeneration — the starting cells were reprogrammed with episomal plasmids specifically to avoid integration, and the finished sheets were screened to confirm no residual plasmid remained before surgery.

A therapy built on cells carrying permanent, semi-random insertions of cancer-linked genes is not one any regulator will approve. The long march from virus to molecule was, in the end, a march toward cells clean enough to transplant.

iPSCs Reach the Clinic

Take a skin punch or a blood draw from a sick patient. Reprogram those cells into induced pluripotent stem cells (iPSCs) — cells wound all the way back to an embryonic-like state, from which they can become any tissue in the body. Then coax them forward into whatever specialized type you want.

This does two useful things at once.

It lets researchers grow a living piece of a sick person’s body in a culture dish. And it gives them a renewable supply of replacement cells to rebuild what disease has destroyed.

Both payoffs arrive without the step that made human embryonic stem cells (ESCs) so politically fraught: destroying a days-old embryo to harvest them. An iPSC line starts from that skin punch or blood draw. So the supply of pluripotent cells is, in principle, capped only by the number of consenting donors.

Disease in a dish

The first payoff has a nickname: “disease in a dish.”

The iPSCs carry the donor’s complete genome — every mutation that causes their illness included. So when you direct those cells to differentiate (mature into a specialized fate), you get the exact affected cell type, with the exact genetic background, of a specific living patient. Neurons, heart muscle, blood cells — tissue you could never biopsy from a living person — now available in unlimited quantity.

This was pioneered in 2009. Researchers took skin cells from a child with spinal muscular atrophy, an inherited motor-neuron disease, reprogrammed them, and differentiated the iPSCs into spinal motor neurons that visibly degenerated in the dish, recapitulating the disease. Cardiac examples followed quickly. iPSC-derived cardiomyocytes (heart muscle cells) from patients with long-QT syndrome, a genetic rhythm disorder, reproduced the hallmark electrical defect — abnormally prolonged action potentials and a tendency toward arrhythmia — in cells beating on a dish.

The logic extends to neurodegenerative, cardiac, and blood disorders alike. Take the patient’s cells. Make the tissue that fails. Watch it fail under a microscope.

This is exactly why iPSC models matter for drug discovery, where animal models routinely mislead because a mouse neuron and a human neuron are not the same thing. More recently, one group pushed this to scale — an iPSC library from 100 people with sporadic (non-inherited) amyotrophic lateral sclerosis, or ALS, every one of them differentiated into motor neurons that reproduced the disease’s signatures: reduced survival, accelerated breakdown of the neurites that wire neurons together, and disordered gene expression. Then they screened that population of cells against drugs that had already been tried in ALS trials. The verdict was sobering: 97% had failed to protect the neurons. But the screen also pointed to a promising new combination — baricitinib, memantine, and riluzole. An earlier ALS study found that motor neurons from patients were pathologically over-excitable, and that a potassium-channel drug, retigabine, quieted them. That finding went directly into a human clinical trial.

iPSC-derived cardiomyocytes are now a standard tool in pharmaceutical toxicity testing, used to flag drugs that dangerously lengthen the QT interval before those compounds ever reach people. And in the most personalized version of the idea, a patient’s own iPSC-derived cells can be used to test which drug, or which dose, their particular biology will tolerate.

Replacing what’s lost

The second payoff is harder: transplantation. Growing healthy cells to replace the ones the body has lost.

The founding milestone came on September 12, 2014, when an ophthalmologist-scientist, Masayo Takahashi, and her team at Japan’s RIKEN institute performed the world’s first transplant of iPSC-derived cells into a human. The patient was a woman in her seventies with the “wet” form of age-related macular degeneration. She received a sheet of retinal pigment epithelium — the support-cell layer behind the light-sensing retina, whose failure drives the disease — grown from her own reprogrammed skin cells. The graft was autologous, meaning patient-derived. Four years on, the transplanted sheet had survived beneath her retina with no tumor and no adverse events. Her vision wasn’t restored, but it had stopped deteriorating.

But that trial also delivered an early warning. When the team prepared a second patient, that person’s iPSCs turned out to have acquired six new mutations not present in the original skin cells — at least one in a site of concern. A vivid reminder that reprogramming, and the long cell culture it requires, can introduce genomic changes. The program was paused. When it resumed, it switched away from the slow, bespoke autologous model toward banked donor cells.

Since then the pipeline has broadened to the body’s other irreplaceable cell types. The most advanced program targets Parkinson’s disease, where the brain loses the dopaminergic neurons (dopamine-producing nerve cells) that control movement. A team led by Jun Takahashi at Kyoto University’s Center for iPS Cell Research and Application differentiated iPSCs into dopaminergic progenitor cells, purified them using an antibody against a marker called CORIN to weed out unwanted cell types, and transplanted them into the brains of seven patients aged 50 to 69. The results, published in Nature in 2025, reported no tumors and no serious adverse events over two years. On the standard Parkinson’s motor scale, the MDS-UPDRS, symptoms during “off” periods — when medication has worn off — improved by an average of 20.4%, and during “on” periods by 35.7%. On the strength of that trial, Japanese regulators granted Sumitomo Pharma’s product — AMCHEPRY (raguneprocel) — a conditional, time-limited approval on March 6, 2026. That makes it one of the world’s first approved therapeutics derived from iPS cells. A national health panel approved insurance coverage that May.

Diabetes is the second major front. Researchers have worked out how to march pluripotent stem cells through the developmental steps that produce pancreatic islet cells, including the insulin-producing beta cells that type 1 diabetes destroys. The leading clinical product, Vertex Pharmaceuticals’ zimislecel (VX-880), is striking. In its trial, all 12 patients given a full dose of the transplanted islets were producing glucose-responsive insulin by day 90, and all 12 reduced or eliminated their injected insulin, with 10 of 12 reaching full insulin independence. Notice, though, that zimislecel is derived not from iPSCs but from embryonic stem cells, and it is allogeneic — made from donor cells, not the patient’s own — so recipients have to take lifelong immunosuppressant drugs to prevent rejection. The iPSC-specific milestone came from China in 2024, where a woman with type 1 diabetes received islet cells grown from her own reprogrammed cells. She became insulin-independent from day 75, and the fraction of the day her blood sugar stayed in the healthy range climbed from 43% to 96%. Because the graft was autologous, she needed no anti-rejection drugs for it. In the heart, the company Heartseed — spun out of work by Keiichi Fukuda at Keio University — is running an early-phase Japanese trial injecting spheres of allogeneic iPSC-derived ventricular cardiomyocytes into the failing hearts of patients with advanced heart disease.

Off-the-shelf or made-to-order

Running through all of these programs is a single strategic fork.

Autologous therapies, made from the patient’s own cells, are a genetic match and provoke no immune rejection. That spares the patient toxic immunosuppression — the great advantage on display in the diabetes and original eye cases. But they are slow and expensive. Each patient’s line has to be derived, expanded, differentiated, and quality-checked over many months, at a cost that can run into hundreds of thousands of dollars. You cannot stockpile it, and you cannot scale it.

The allogeneic alternative treats iPSCs as an “off-the-shelf” product. The trick is to bank cells from carefully chosen donors who are homozygous for the human leukocyte antigens, or HLA — the cell-surface proteins the immune system reads to tell “self” from “foreign.” A donor carrying two identical copies of common HLA types can immunologically match a large slice of the population. Kyoto’s iPS Cell Stock has built lines from seven such donors that match roughly 40% of Japanese people, and they have already supplied more than ten clinical trials. Off-the-shelf cells are faster, far cheaper per patient, and can be manufactured in advance. But recipients are not perfect matches, so most allogeneic protocols still require some immunosuppression.

This is where the clean ethical advantage of iPSCs gives way to thornier questions.

Sidestepping embryo destruction defused the objection that had dogged ESC research for a decade. But as zimislecel shows, embryo-derived cells haven’t vanished from the clinic — so the advantage is real, just not yet universal. And in place of the old objection sit new ones.

The foremost is safety. Any residual undifferentiated iPSC left in a graft can form a teratoma, a tumor containing a chaotic mix of tissue types. This is why programs invest so heavily in purification, purity testing, and post-transplant MRI surveillance — and why those acquired mutations in the eye trial mattered so much. Regulators have their own dilemma, visible in Japan’s conditional-approval pathway, which lets cell therapies reach patients before efficacy is fully proven, trading speed against certainty.

And equity looms over all of it. Therapies that cost six or seven figures per patient will not be evenly accessible. Worse, HLA-matched banks built around the commonest haplotypes in one population systematically serve majority-ancestry patients better than people of minority or mixed heritage. Left unaddressed, the off-the-shelf solution could widen the very gaps that personalized medicine promises to close.

Rewinding Age, Not Identity

In 2006, Shinya Yamanaka found four transcription factors — Oct4, Sox2, Klf4, and c-Myc, collectively OSKM — powerful enough to walk an ordinary adult cell all the way back to an embryonic state. Run them long enough and a skin fibroblast becomes an induced pluripotent stem cell: a blank slate that can become any tissue, but one that has forgotten what it used to be.

That total erasure is exactly the problem for anyone hoping to use reprogramming on a living body. Turn a pluripotent cell loose in an adult mouse and it doesn’t rebuild an organ. It grows a teratoma — a grotesque tumor packed with a disordered jumble of tissues, hair and teeth and gut lining sprouting where none belong.

For most of two decades, this confined reprogramming to the culture dish.

But somewhere along that road from adult to embryo, something useful happens before identity is lost. As the factors run, a cell doesn’t only forget what it is. It also gets measurably younger, shedding the molecular wear that separates an old cell from a young one.

The frontier insight that turned Yamanaka’s stem-cell trick into a longevity obsession is that these two effects can be pried apart: capture the rejuvenation, but stop short of the erasure, by switching the factors on briefly and then off again before the cell passes the point of no return. This is partial, or transient, reprogramming.

What the factors act on is the epigenome — the layer of chemical tags sitting on DNA and on the histone proteins that DNA winds around, which dictates which genes a cell switches on or off without touching the underlying genetic sequence. Two cells in your body can carry identical DNA — a neuron and a liver cell — and differ entirely in their epigenome.

Over a lifetime this layer drifts and degrades. And that drift is much of what aging is at the cellular level.

Reprogramming factors are what biologists call pioneer factors: proteins that can bind DNA even where it’s densely packed and silenced, prying the chromatin open and resetting those tags. Cycled on and off, they begin rewriting the epigenome toward a youthful configuration — but get withdrawn before they finish the job of converting the cell’s identity.

The proof that this works inside a living animal came in 2016, from Alejandro Ocampo, Juan Carlos Izpisua Belmonte, and colleagues at the Salk Institute, published in Cell. They engineered mice whose cells carried the four OSKM genes under the control of doxycycline, a drug they could add to the animals’ water to switch the factors on at will.

But here’s the catch. Continuous expression was lethal — it killed the mice within days and seeded tumors, the teratoma danger made real. So they pulsed it: two days on, five days off, repeated in cycles, never letting any cell drift too far.

They first tested this in progeroid mice — animals carrying the LMNA mutation that causes Hutchinson-Gilford progeria, the human premature-aging syndrome in which children age rapidly and die young. Cyclic reprogramming raised the median lifespan of these mice by roughly a third and their maximum lifespan by about a fifth, while reversing concrete hallmarks of aging in their tissues: fewer DNA double-strand breaks, restored heterochromatin (the tightly packed, silenced regions of the genome), reduced cellular senescence (cells that have stopped dividing but linger and inflame their surroundings), and healthier mitochondria.

Just as important, in normal, naturally aged mice the same pulsed treatment improved the regeneration of injured pancreas and muscle. The benefit wasn’t a quirk of the progeria model. It was a genuine rejuvenation of ordinary old tissue.

Reading the clock backward

To claim a cell has grown younger, you need a way to measure cellular age. The tool that made the whole field quantitative is the epigenetic clock.

Its raw material is DNA methylation — the attachment of small methyl groups onto cytosine bases, almost always where a cytosine sits next to a guanine, a so-called CpG site. Methylation generally tamps a gene down, and the pattern across the genome shifts with age in a strikingly regular way. In 2013, Steve Horvath at UCLA showed that you could read the methylation state of just 353 carefully chosen CpG sites, feed them into a formula, and estimate a tissue’s age to within a few years. And remarkably, the same clock worked across nearly every human tissue and cell type — from blood to brain to breast.

The Horvath clock and its successors turned “biological age” from a vague notion into a number. And that number does something a genetic readout cannot: it runs backward under reprogramming. Push a cell all the way to pluripotency and its methylation age resets essentially to zero — Horvath noticed this himself. The bet of partial reprogramming is to wind that clock back part of the way, and stop.

That precision is exactly what the next landmark exploited. In 2020, Yuancheng Lu, David Sinclair, and collaborators at Harvard, writing in Nature, narrowed the cocktail. They dropped c-Myc — the factor most associated with cancer — and used only three, Oct4, Sox2, and Klf4 (OSK), reasoning that the safer trio could rejuvenate without the oncogenic risk. They delivered the genes to the eye with an adeno-associated virus, a harmless engineered virus used as a gene-therapy vehicle, again under an inducible switch so expression could be shut off.

In mice whose optic nerves had been crushed, OSK roughly doubled the survival of retinal ganglion cells — the neurons that carry vision from eye to brain — and produced a severalfold increase in regenerating nerve fibers, which adult central neurons essentially never regrow on their own. The effect extended to age and disease: OSK restored lost vision in twelve-month-old mice and in animals with a glaucoma-like condition driven by elevated eye pressure, recovering roughly half the lost visual acuity.

And the mechanism tied directly back to the clock. Injury and age had driven the neurons’ DNA methylation in an “old” direction, and OSK reversed it — an effect that vanished when the researchers knocked down TET1 and TET2, the enzymes that actively strip methyl groups off DNA. Reprogramming wasn’t scrambling the methylation pattern. It was guiding it back toward a youthful state. And a year of treatment produced no detectable tumors.

Age and identity come apart

The deeper conceptual payoff of this work is the decoupling: biological age can reset before cellular identity is lost. These are not the same process running on the same timer. A neuron rejuvenated by a few weeks of OSK is still unmistakably a neuron. A treated fibroblast is still a fibroblast.

The clearest demonstration came in 2022 from Wolf Reik’s group at the Babraham Institute. They developed what they called “maturation phase transient reprogramming” — running the factors in human skin cells for about thirteen days, long enough that the cells transiently lost their fibroblast identity, then withdrawing the factors and letting them snap back. The cells re-emerged as fibroblasts, but ones whose methylation and transcriptional clocks read roughly thirty years younger, and which behaved younger too: producing more collagen, migrating into wounds more readily.

Identity was recovered. Age was not.

This separability is the field’s central wager: that dedifferentiation (becoming a less specialized cell) and rejuvenation (becoming a younger cell) are distinct trajectories that merely happen to travel together during full reprogramming, and that the second can be harvested without committing to the first.

Nature already runs this reset

None of this would be plausible if aging were a one-way accumulation of irreversible damage. Like rust.

The strongest hint that it isn’t comes from the most ordinary fact in biology: every generation starts young. An aged parent produces an offspring whose biological clock reads near zero, which means the information needed to specify a youthful cell is not destroyed by a lifetime of living. It’s recoverable.

In 2021, Csaba Kerepesi, Vadim Gladyshev, and colleagues sharpened this by tracking epigenetic age through early development, and found that biological age does not simply start low and climb. The germ cells that fuse at conception are themselves aged, yet the early embryo actively drives its epigenetic age downward, reaching a minimum — a “ground zero” — around the time of gastrulation and early implantation, after which organismal aging begins to count up. Nature, in other words, already runs a rejuvenation program once per generation, resetting the clock as the new embryo forms. Partial reprogramming is an attempt to invoke a controlled fraction of that same natural reset in cells that have already differentiated and aged.

This reframes what aging fundamentally is. In the information theory of aging, articulated most prominently by Sinclair, the genome is digital — a stable sequence of four bases, faithfully copied and largely intact even in old cells — while the epigenome is analog, a pattern of marks that degrades as it is read, copied, and buffeted by stress and damage. Aging, on this view, is not primarily the corruption of the genetic text but the gradual loss of the epigenetic information that tells each cell which parts of that text to use. The supporting experiment is pointed: when Sinclair’s group induced repeated DNA breaks that cells dutifully repaired — disrupting the epigenome while leaving the DNA sequence intact — the mice aged faster by multiple measures, including their methylation clocks, and OSK could push those measures back.

If aging is lost information rather than broken hardware, then a backup copy must persist somewhere for reprogramming to read from, and the recoverability of youth becomes a question of access rather than reconstruction.

That is the proposition the rest of the field is now racing to test: that what time takes from a cell is not its identity or its genome, but the legibility of the instructions it has carried all along.

The Dedifferentiation Tightrope

Reprogramming runs a cell’s developmental history in reverse. The same four Yamanaka factors that turn a skin cell into a pluripotent stem cell — one capable of becoming any tissue in the body — also strip away whatever made it a skin cell in the first place.

And that’s exactly the problem with using reprogramming as a medicine inside a living animal.

A liver is a liver because its cells cling to their identity. A body whose cells forget what they are isn’t getting younger. It’s dissolving.

The therapeutic promise and the central danger spring from the same molecular act. Only a narrow margin separates them.

The Cancer Cliff

The clearest demonstration of that danger came in 2013. Manuel Serrano’s group engineered mice carrying the four Yamanaka factors — Oct4, Sox2, Klf4, and c-Myc, or OSKM for short — behind a drug-activated switch, so the factors could be turned on throughout the body.

When they were, the animals didn’t grow younger. They grew teratomas.

Teratomas are grotesque tumors — a chaotic jumble of tissue types all growing where they don’t belong. Hair. Gut lining. Muscle. Glandular cells. They’re the signature of pluripotent cells loose in an adult body, because a cell that has reverted all the way to an embryonic-like state stops taking cues from its neighbors and instead tries to build tissues at random. Examining the stomach, intestine, pancreas, and kidney, the researchers found clusters of dedifferentiated cells — cells that had slid back from their specialized form toward an unspecialized one — switching on NANOG, the protein that marks full pluripotency. Reprogrammed stem cells were even circulating in the animals’ blood.

In vivo reprogramming worked. That was exactly the trouble.

This isn’t a bug you can engineer away, because dedifferentiation and cancer run on overlapping machinery. A defining feature of many aggressive tumors is precisely that their cells have shed their mature identity and reverted to a primitive, fast-dividing state — the exact direction reprogramming pushes them. One of the four factors, c-Myc, is a classic oncogene — a gene that drives cancer when overactive — and among the most frequently amplified genes in human tumors. The other three aren’t innocent either. Each governs the self-renewal programs that cancers routinely hijack.

Nudge a cell partway back down its developmental path and you may rejuvenate it. Push it too far and you’ve manufactured something indistinguishable from a tumor cell.

The cliff is steep. Leave OSKM switched on continuously and mice don’t last long — in some models median survival is about five days, the animals dying of liver and intestinal failure as tissues lose their structure and cells proliferate without restraint. The line between winding back the clock and triggering runaway growth is measured in days of exposure, not weeks.

Dose, Timing, and Target

The escape is to never let reprogramming finish.

In 2016, Juan Carlos Izpisúa Belmonte’s group at the Salk Institute pulsed the factors — on for two days, off for five, week after week — in mice with progeria, a premature-aging disease caused here by a mutation in the lamin A gene. This cyclic, “partial” reprogramming reset the hallmarks of aging without carrying cells all the way to pluripotency. Spinal curvature, skin, and cardiovascular function improved. Median lifespan rose by roughly a third, maximum lifespan by about 18 percent. No teratomas.

The lesson hardened into the field’s governing principle: rejuvenation and tumor formation can be uncoupled, but only by tightly controlling dose, duration, and location. Expression has to be transient, ideally cyclic, and confined to the tissue that needs it.

Too little does nothing. Too much, or for too long, kills.

The therapeutic window is real but narrow. And where its edges lie almost certainly differs from one tissue to the next.

Delivery Into a Living Body

Controlling reprogramming in a petri dish is one thing. Doing it inside a person is another, and the delivery problem is brutal.

The workhorse of gene therapy is the adeno-associated virus, or AAV — a hollowed-out, non-disease-causing virus repurposed to ferry genes into cells. But AAV holds only about 4.7 kilobases of cargo, and the four Yamanaka factors plus the regulatory sequences needed to control them don’t fit. So researchers split the payload across two or three viruses and hope each cell receives the full set, or they drop a factor altogether.

The widely used three-factor cocktail, OSK, drops c-Myc — both to shrink the cargo and to remove the most dangerous oncogene. David Sinclair’s group used exactly this combination, delivered by AAV into retinal cells, to restore sight in aged and glaucoma-damaged mice, and reported no tumors even after fifteen months of expression in those non-dividing neurons. That reassurance may hold precisely because mature neurons don’t divide, and so can’t easily spawn a tumor.

Then there’s control. The factors have to sit behind an inducible switch — a genetic on/off tripped by a drug like doxycycline or tamoxifen — so a clinician can pulse them rather than leave them blaring. And AAV doesn’t respect tissue boundaries. Virus aimed at one organ leaks into others, which raises the prospect of reprogramming firing where it was never wanted — in cells that, unlike neurons, divide readily.

What We Still Don’t Know

Suppose the mouse results are exactly as advertised. Large unknowns still stand between them and a human therapy.

The first is durability. Reprogramming is usually judged by the epigenetic clock — an estimate of biological age read from DNA methylation, the pattern of small chemical tags that sit on genes and tune them up or down without altering the underlying sequence. A pulse of factors winds that clock backward. But whether the rejuvenated state holds after the factors switch off, or simply drifts back toward old, is unsettled. And whether resetting the clock actually reverses aging — rather than merely moving a marker that happens to correlate with it — is itself debated.

The second is systemic safety. A treatment that helps one tissue might quietly seed a tumor in another years later. Aging is a whole-body process, yet nearly every success so far has been confined to a single organ.

The third, and the largest, is the species gap. Mice are short-lived and tumor-prone in ways that may flatter what reprogramming does or distort it. Their cells reprogram more readily than human cells. And a third more lifespan in a progeria mouse is a long way from safe rejuvenation in a person who then has to live for decades, cancer-free, after treatment.

Those are the stakes that have drawn billions of dollars. Altos Labs launched in 2022 with $3 billion in funding and Yamanaka himself as an adviser, and rival firms are pressing toward early human trials.

They’re also why no one can yet say whether this particular tightrope can be walked inside a human body at all.

Where the Capital Is

The premise behind all this money is simple. In 2006, Shinya Yamanaka showed that just four transcription factors — proteins that switch genes on and off, abbreviated OSKM for Oct4, Sox2, Klf4, and c-Myc — could wind an ordinary adult cell all the way back to an embryonic-like stem cell. The implication is bigger than it sounds. A cell’s age isn’t only accumulated damage. It’s partly an information state: a pattern of chemical marks on DNA and its packaging proteins, called the epigenome, that decides which genes are switched on.

And patterns can be rewritten.

A decade later, Juan Carlos Izpisúa Belmonte’s lab showed that applying those factors briefly and repeatedly — “partial” reprogramming, stopping well short of turning the cell into a stem cell — rejuvenated tissues and extended lifespan in fast-aging mice. The bet now consuming billions of dollars is that this rollback can be made strong enough to matter, in the right cells, without erasing a cell’s identity or tipping it into cancer.

The money is already on the table. Here’s who put it down, and what each is betting.

Altos Labs

The biggest single bet is Altos Labs. It launched in January 2022 with roughly $3 billion — the best-funded debut in biotech history — from backers including Jeff Bezos and Yuri Milner. Yamanaka himself sits on its scientific advisory board, unpaid.

Altos went on a hiring spree. It brought in Belmonte, reprogramming biologist Wolf Reik, and Steve Horvath — who built the first “epigenetic clocks” that estimate a cell’s age from its chemical marks — into three Institutes of Science in the San Francisco Bay Area, San Diego, and Cambridge, England. Roughly twenty principal investigators, run by former GlaxoSmithKline R&D head Hal Barron and ex–National Cancer Institute director Rick Klausner.

And here’s the striking part. Altos refuses to promise products. It calls its work “cellular rejuvenation programming” and is structured as a set of richly funded academic-style labs. The bet: deep mechanistic understanding must come before any medicine.

Retro Biosciences

Retro Biosciences is the opposite temperament. Founded in 2021, its entire $180 million seed round came from a single person: Sam Altman, who disclosed in 2023 that he’d personally written the check.

Retro’s goal is to add ten years to “healthspan” — years lived in good health, not raw years alive. But notice that partial reprogramming is only one of five discovery programs, spread across three approaches: cellular reprogramming, autophagy (the cell’s system for clearing and recycling its own waste), and plasma-inspired therapeutics. The company became clinical-stage in December 2025 — but with an autophagy pill, RTR242, not a reprogramming drug. Its reprogramming work is still preclinical. Retro has since gone after a far larger raise, reported at around $1 billion.

Its bet is diversification. Reprogramming may pay off, but Retro is hedged against several mechanisms of aging at once.

NewLimit

NewLimit, founded in 2021 by Coinbase CEO Brian Armstrong with investor Blake Byers and stem-cell scientist Jacob Kimmel, took the more focused, data-driven path. The founders seeded it with about $105 million of their own money. A $40 million Series A followed in 2023, then a $130 million Series B in 2025 led by Kleiner Perkins, plus another $45 million from Eli Lilly and others — total funding near $350 million at a valuation around $1.6 billion.

NewLimit doesn’t rely on the classic four factors. Instead, it screens many transcription-factor combinations to find which ones make a specific cell type act younger, then delivers them as mRNA — short-lived genetic instructions telling a cell to make a protein — packaged in lipid nanoparticles, tiny fat bubbles. Its lead program rejuvenates liver cells (hepatocytes), with programs in immune T cells and blood-vessel lining behind it. First human trial: within roughly two years.

Life Biosciences

Life Biosciences, co-founded by Harvard’s David Sinclair, has moved fastest toward actual patients — by starting in the eye. Its Partial Epigenetic Reprogramming platform uses only three factors, OSK, dropping c-Myc because of its association with cancer.

On January 28, 2026, the FDA cleared the company’s Investigational New Drug application — the filing that grants permission to begin human testing — for ER-100. This is the first in vivo reprogramming therapy, ie, delivered into the living body rather than to cells in a dish, cleared for clinical trials anywhere. The Phase 1 study treats optic neuropathies, including open-angle glaucoma and a form of optic-nerve stroke called NAION.

Starting in the eye is shrewd. It’s small, walled off from the immune system, and easy to measure — you simply test whether vision improves. A liver candidate, ER-300, aimed at the fatty-liver disease MASH, is next in line.

Calico

Calico is the patient, well-capitalized incumbent. Founded by Google in 2013 under former Genentech chief Arthur Levinson, its partnership with AbbVie drew some $1.75 billion. But its bet was the basic biology of aging broadly — naked mole rats, computational target-hunting — rather than reprogramming specifically. That long arc met a hard limit in November 2025, when AbbVie ended the collaboration after an ALS drug failed in trials.

Turn Biotechnologies

Turn Biotechnologies sits at the other extreme: a smaller, platform-focused player. Its ERA technology — Epigenetic Reprogramming of Aging, licensed from Stanford — delivers reprogramming factors as mRNA transiently, halting before a “point of no return” so cells keep their identity. Lead work is in skin, immune cells, and osteoarthritis.

What They Agree On, and What They Don’t

What unites all of them is one conviction: aging is, in meaningful part, a reversible epigenetic state.

Where they diverge is on the variables nobody has settled:

• Which factors to use — the classic OSKM, the safer OSK, or bespoke cell-specific sets.
• How to deliver and switch them off — transient mRNA, gene-therapy-style in vivo systems, or reprogramming cells outside the body and returning them.
• Which organ to attempt first.
• How loudly to promise results.

So you get three camps. Altos and Calico bet that understanding precedes medicine. Life Biosciences and NewLimit bet they can get a controllable therapy into clinics now. Retro bets across several aging mechanisms at once.

But all of them are making the same wager, still unproven in humans: that a reset achievable in a dish or a mouse will hold in a person — without the cell forgetting what it was meant to be.

The Five-Year Pipeline

In vivo reprogramming is reaching the clinic. But not all at once — one organ at a time.

And the order isn’t random. It boils down to a single practical question: where can you deliver the factors, switch them on, contain them, and watch them work?

By that test, the eye wins. It’s already produced the first human trial.

The eye goes first

The eye is an ideal proving ground, for three reasons.

First, it’s immunologically privileged. The retina sits behind a blood–tissue barrier that tolerates foreign proteins the rest of the body would attack. This matters more than it sounds, because the delivery vehicle itself — adeno-associated virus, or AAV, a small, non-disease-causing virus engineered to ferry genes into cells — normally provokes immune responses that can shut a therapy down.

Second, it’s anatomically closed. A single intravitreal injection — a shot into the gel-filled cavity of the eyeball — bathes the retina with a minute dose while keeping the virus from spreading body-wide.

Third, it’s transparent. You can image the treated cells and measure their electrical output without ever cutting anyone open.

The lead candidate is Life Biosciences’ ER-100. It carries OSK — OCT4, SOX2, and KLF4, three of Yamanaka’s four factors. The fourth, c-Myc, is a well-known oncogene — a gene that can drive cancer — and it’s left out on purpose.

The OSK genes sit under a doxycycline-inducible switch. They stay silent until the patient swallows the antibiotic doxycycline, which flips them on, and go silent again the moment the drug is stopped.

The science traces back to a 2020 Nature paper out of David Sinclair’s Harvard lab. (Sinclair co-founded the company.) That paper showed something remarkable. OSK delivered to retinal ganglion cells — the neurons whose axons form the optic nerve — restored a youthful pattern of DNA methylation (the chemical methyl tags that silence genes without touching the underlying sequence), regrew crushed optic-nerve fibers roughly fivefold, and reversed vision loss in aged mice and in a glaucoma model.

Notice that the effect required the demethylase enzymes TET1 and TET2 — the enzymes that strip those tags off. In other words, the cells were recovering old regulatory information, not rewriting their genome.

In monkeys with an injury mimicking NAION — non-arteritic anterior ischemic optic neuropathy, essentially a stroke of the optic nerve, sudden and currently untreatable — ER-100 plus doxycycline restored both methylation patterns and retinal ganglion cell function, in prevention and rescue settings alike.

The FDA cleared the trial on January 28, 2026. That makes ER-100 the first reprogramming-based rejuvenation therapy to reach humans. The first ever.

The Phase 1 study (NCT07290244) is deliberately small and cautious. Up to 18 patients — 12 with open-angle glaucoma, 6 with NAION — each gets a single injection in one eye, at one of two doses (2 or 6 × 10¹¹ viral genomes), followed by an eight-week course of doxycycline. One sentinel patient is dosed and watched for 28 days before anyone else follows. Safety is monitored through Year 5. The endpoints are safety, tolerability, and immune response first; vision second.

Liver next, then the hard organs

The liver is the logical second target, because it offers what the eye offers — clean delivery — just by a different route.

The AAV8 serotype’s protein shell naturally homes to hepatocytes, i.e., liver cells. The property is called hepatotropism, and it means an intravenous infusion concentrates the virus in the liver almost automatically.

Life Biosciences’ ER-300 runs the same OSK platform at MASH — metabolic dysfunction-associated steatohepatitis, formerly NASH — fatty liver that’s progressed to inflammation and scarring. Millions of people have it, and good drugs are scarce. In a mouse MASH model, ER-300 significantly improved liver biomarkers: the enzymes ALT and AST, total cholesterol, bile acids, and NAFLD score. That data was presented at the Aging Research and Drug Discovery (ARDD) meeting in 2025.

But make no mistake, it’s still preclinical — years behind ER-100.

Heart and brain are harder, and their timelines longer. The reason is delivery: neither organ can be flushed with virus and sealed off the way the eye can, and the brain throws in the blood–brain barrier on top.

The biology, though, is encouraging.

A 2021 Science study showed that transient, heart-specific OSKM expression pushed adult cardiomyocytes — muscle cells that normally almost never divide — back toward a fetal-like, proliferative state, regenerating tissue after a heart attack. Sustained expression, on the other hand, produced heart tumors. Which underscores the whole point: control is everything.

In the brain, briefly switching OSK on in the memory-encoding hippocampal neurons of aged mice restored recall toward youthful levels.

Both are pipeline programs, not clinical candidates. Realistically, they sit beyond the five-year window.

The chemical alternative

Everything above is gene therapy — a virus carrying DNA. Chemical reprogramming is a fundamentally different bet: induce the same change with small molecules. No virus, no new genes.

Hongkui Deng’s group at Peking University proved it’s possible. In 2013 they generated chemically induced pluripotent stem cells — CiPSCs — from mouse cells, using a cocktail of seven small molecules, at about 0.2% efficiency. The point being, the Yamanaka transcription factors weren’t strictly required after all. In 2022 they pulled it off in adult human cells, converting human fibroblasts to pluripotency — the capacity to become any cell type — through a stepwise chemical protocol. That was the first time anyone had done it in human cells with molecules alone.

The clinical appeal is concrete. Small molecules don’t insert into the genome. Their dose is easy to titrate. And unlike a virus confined to one organ, a well-designed compound could in principle be swallowed as a pill and reach tissues throughout the body.

The translational goal is partial chemical reprogramming: a cocktail that nudges cells toward a younger state and gets withdrawn before they hit pluripotency. Sinclair’s lab reported six such cocktails in 2023 that reversed molecular markers of age in cultured cells within a week, without triggering a stem-cell state — though the work drew skepticism about how robust it really is.

No oral rejuvenation molecule is in trials yet. Chemically, the field sits roughly where viral reprogramming did a decade ago.

The regulatory shape

Three features define the path to approval.

First, the safety switch. A reprogramming therapy you can turn off is the central argument for letting one into a human body in the first place. That’s why the doxycycline-inducible design is doing as much regulatory work as biological work.

Second, cyclic dosing — an idea Ocampo and colleagues established back in 2016. Continuous OSKM expression killed mice outright, by causing teratomas: tumors of jumbled tissue types, the signature of a cell that’s gone fully pluripotent in the wrong place. But short pulses — two days on, five off — in fast-aging progeria mice eased the hallmarks of aging and extended median lifespan by a third. Partial, pulsed, reversible. ER-100’s eight-week course inherits this logic directly.

Third, endpoints. Aging isn’t an FDA-recognized disease, so these therapies have to enter through specific conditions — glaucoma, MASH — measured with conventional yardsticks like vision and liver enzymes. The longer-term ambition is to use epigenetic clocks — algorithms that estimate biological age from DNA methylation levels at specific genomic sites; the original 2013 clock reads 353 of them — as biomarkers of rejuvenation itself. But the FDA hasn’t accepted any such clock as a surrogate endpoint. It wants to know first that lowering a clock score actually lowers disease risk. Until it does, “we made the cells younger” doesn’t earn approval on its own.

What five years will settle

The next five years will settle three questions that mouse data simply can’t.

Safety in humans. Does OSK provoke tumors or loss of cell identity in people, and is the doxycycline switch tight enough to stop it? ER-100 is the first real test.

Durability. How long does a single course last, and does the benefit require repeated cycling?

Generalization. Indeed, nearly every dramatic result so far is in mice — often progeria mice, engineered to age abnormally fast — with primate data only beginning to bridge the gap.

By 2031, the human trials now starting should tell us whether the rejuvenation phenotype is a deep property of mammalian biology, or just an artifact of the animals it was discovered in.

The Long Bet

The reprogramming toolkit we have today was never built for rejuvenation.

When Shinya Yamanaka went hunting for the genes that could turn an adult cell back into an embryonic-like one, he wasn’t after youth. He was after pluripotency — the capacity to become any cell type. And the four transcription factors he found — Oct4, Sox2, Klf4, and c-Myc, or OSKM for short — were optimized for one destructive job: erasing a cell’s identity. The last of them, c-Myc, is a notorious cancer-promoting gene, which is why most rejuvenation work now throws it out and uses just the first three, OSK. Partial reprogramming borrows the same factors but stops the process early. The goal is to wind back the molecular marks of age — the chemical tags on DNA and the proteins that package it, which control which genes are switched on without touching the DNA sequence itself — while keeping the cell as what it is.

It’s using a demolition crew for a renovation.

The long bet starts with the premise that better tools exist. And the factor sets people are chasing now are increasingly machine-discovered, rather than scavenged from the pluripotency literature. NewLimit, founded by Coinbase’s Brian Armstrong, runs a “lab in a loop”: computational models predict candidate combinations, high-throughput experiments test them, and the results retrain the model. Shift Bioscience used a cellular-age clock and active machine learning to nominate SB000, a single-gene target it claims rejuvenates cells without ever switching on the dangerous pluripotency program. The aim is to decouple “younger” from “less itself” — to find interventions selected specifically for restoring youthful function, not inherited from a recipe for making stem cells.

The further endpoint is to drop gene delivery altogether. Genes are awkward medicine. They have to be packaged into viruses, they’re hard to dose precisely, and once they’re inside your body you can’t easily switch them off. By contrast, small molecules — ordinary drugs you swallow — are tunable, reversible, and cheap to make. The proof that chemistry can do the work of transcription factors landed in 2022, when Deng Hongkui’s group at Peking University reported in Nature that a cocktail of small molecules alone could reprogram adult human cells all the way to pluripotency, with no Yamanaka genes at all. In 2023, a Harvard group including David Sinclair reported in the journal Aging that six small-molecule combinations reversed measures of cellular age in cultured human cells within a week. It’s early, it’s contested, and it was run in dishes rather than animals. But it’s a gesture toward the destination.

And that destination is a pill. A periodic epigenetic reset, taken perhaps once a year like a flu shot, that nudges aging cells back toward a younger state without ever touching the genome.

Whether reprogramming ever reaches the whole body is the harder question, and it’s fundamentally an engineering problem, not a biological one.

The animal proofs are solid. A 2016 study cyclically switched OSKM on and off in mice with a premature-aging syndrome and extended their lifespans. The landmark 2020 Nature paper delivered OSK to the eye and restored vision in glaucomatous and aged mice while resetting their DNA methylation age — the pattern of methyl tags on DNA that an “epigenetic clock,” an algorithm reading those tags at specific sites, uses to estimate biological age. And in 2024 a Rejuvenate Bio team delivered OSK systemically with an adeno-associated virus — AAV, a hollowed-out virus used as a gene-therapy vehicle — to 124-week-old mice and more than doubled their remaining median lifespan.

But three problems separate mouse from human.

Delivery. AAV reaches some tissues and not others, provokes immune responses, and generally can’t be re-dosed, because the body learns to neutralize it.

Dose and duration. Push reprogramming too far and cells lose their identity or form teratomas — chaotic tumors of mixed tissue types. So the “partial” window is narrow, and it differs by tissue.

Control. The mouse experiments rely on a genetic switch toggled by the antibiotic doxycycline, a precision no approved human therapy can yet match.

The engineering answers being explored are inducible, reversible switches; orally dosed small molecules whose effect stops when you stop taking them; engineered viral capsids and lipid-nanoparticle mRNA for targeted delivery; and the safer, AI-designed factor sets from above. The first clinical beachhead reflects an obvious strategy — start where these problems are smallest. In January 2026 the FDA cleared Life Biosciences’ ER-100 — an AAV carrying OSK, injected into the eye and gated by oral doxycycline — for a Phase 1 trial in optic neuropathies, the first in-human test of cellular reprogramming. The eye was chosen precisely because it’s contained, immune-privileged, and injectable.

Whole-body treatment lies far past it.

Reprogramming won’t arrive alone. The long bet assumes it becomes one layer of an integrated stack, not a standalone cure. Aging degrades cells in several distinct ways, and different interventions target different layers. Senolytics — drugs like the dasatinib-plus-quercetin combination that selectively kill senescent cells, the worn-out cells that stop dividing but refuse to die and instead leak inflammatory signals into the surrounding tissue — clear away damage that reprogramming can’t repair. Raising NAD+, the metabolic coenzyme every cell needs to produce energy and to power a family of repair enzymes called sirtuins, aims to restore the fuel supply for that repair, though human data on precursors like NMN remain mixed. Reprogramming itself rewinds the epigenetic clock. The thread tying all of this together is measurement: the same DNA-methylation clocks — Steve Horvath’s original 2013 version, and mortality-trained successors like GrimAge — that estimate biological age could become the shared readout that lets these interventions be compared, sequenced, and combined.

For any of this to become real medicine instead of billionaire-funded hope — Altos Labs launched in 2022 with $3 billion and Yamanaka as an adviser; Retro Biosciences sought $1 billion in 2025 — three things have to come true.

Safety at scale. Reprogramming has to reverse aging markers across many tissues, durably, without ever tipping a single cell toward cancer. That bar is far higher in a sixty-year-old than in a mouse.

Regulatory acceptance. Today no epigenetic clock is qualified by the FDA as a surrogate endpoint — a measurable stand-in that regulators accept as proof a drug works, instead of waiting years for disease outcomes — and aging itself isn’t even a recognized indication. The TAME trial, Nir Barzilai’s metformin study designed with FDA input to test whether the agency will accept a composite of age-related diseases as a single endpoint, is the field’s deliberate attempt to pry that door open.

Economics. AAV gene therapies are the most expensive drugs ever sold. Hemgenix lists at $3.5 million per dose, Zolgensma at $2.1 million, with manufacturing the dominant cost. A one-time, multimillion-dollar treatment is not population medicine.

Which is exactly why the oral-pill endpoint matters. A small molecule is the only version of this that could ever scale to billions of people.

But the deepest thing Yamanaka changed was conceptual. In 1957 the biologist Conrad Waddington pictured development as a ball rolling downhill through a branching landscape of valleys. Once a cell settled into a valley — became a neuron, a liver cell, a skin cell — it was thought to be stuck there for good.

Fate was a one-way street.

Yamanaka’s factors showed that the ball can roll back uphill, and that the contours of the landscape aren’t carved into the DNA sequence at all — they’re written in the erasable epigenetic layer sitting on top of it. That reframes a cell’s age and identity as something closer to editable software than fixed hardware: a state you can read, write, and reset.

The biology of the core claim is largely settled. You can rewind a cell. What’s left is engineering — delivering the edit to the right cells, controlling its depth, halting it on command, proving it safe, and making it cheap enough to matter. The long bet is simply that these are engineering problems. And that engineering, given enough time, tends to yield.